[BUG]: Can't use PYBIND11_NUMPY_DTYPE with subinterpreters

Required prerequisites

• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker and Discussions to verify that this hasn't already been reported. +1 or comment there if it has.
• Consider asking first in the Gitter chat room or in a Discussion.

What version (or hash if on master) of pybind11 are you using?

3.0.0

Problem description

Registering a DTYPE fails in subinterpreters because the DTYPE is already registered. Any thoughts on how to make this subinterpreter safe?

ImportError: NumPy: dtype is already registered

Edit: I think numpy may not work in subinterpreters; if I suppress this error when it's already loaded, then it looks like numpy throws an error when importing. So it might not be very important to solve until numpy supports subinterpreters.

Reproducible example code

PYBIND11_NUMPY_DTYPE(weighted_sum, value, variance);

Is this a regression? Put the last known working version here if it is.

Not a regression

Start with the PYBIND11_NUMPY_DTYPE(weighted_sum, value, variance) example and reproduce it across subinterpreters. Investigate whether the failure comes from pybind11's dtype registration or NumPy's subinterpreter support; done requires a confirmed safe behavior or a documented blocker.